The US Environmental Protection Agency’s (USEPA) Environmental Monitoring and Assessment Program (EMAP) sampled ∼500 wadeable streams in the Mid-Atlantic Highlands region of the US during the late spring of 1993 to 1995 for a variety of physical, chemical, and biological indicators of environmental condition. Eighty-eight sites that were minimally affected by human activities were chosen to determine the extent to which geographic and stream-based classifications accounted for variation in the composition of riffle macroinvertebrate assemblages. Bray–Curtis similarities among sites were calculated from the relative abundance of macroinvertebrates to assess the strength of classifications based on geography (ecoregions and catchments), habitat (slope and stream order), and water chemistry (conductivity). For comparison, a taxonomic classification (two-way indicator species analysis, TWINSPAN) and a gradient analysis (correspondence analysis, CA) were performed on the macroinvertebrate data. To assess the effect of taxonomic resolution, all analyses were completed at the family level and to lowest practical taxon. The large overall variation within and among ecoregions resulted in a low average classification strength (CS) of ecoregions, although some ecoregions had high CS. Stream order had the highest CS of the habitat and water chemistry classifications. Ecoregion CS increased, however, when stream sites were 1^st stratified by stream order (ecoregions nested within stream order). Nested ecoregion CS did not increase within 1^st-order streams, yet increased within 2^nd- and 3^rd-order streams. CA ordinations and TWINSPAN classification showed a clear gradient of streams along stream size (order), with a clear separation of 1^st- and 3^rd-order streams based on macroinvertebrate composition. The ordinations did not, however, show a distinct clustering of sites on the basis of ecoregions. Overall, the lowest practical taxon level of identification resulted in a clearer pattern of sites in ordination space than did family-level identification, yet only a slight improvement in the different classifications (geographic, habitat, and water chemistry) based on average similarity.